project showcase
5 liveInteractive experiments. Engaging projects.
Agent-based reinforcement learning simulation of Minsky's Financial Instability Hypothesis, testing whether profit-maximising RL agents endogenously generate financial fragility from a stable market equilibrium. Features a full market simulator with leverage, margin calls, and forced liquidation, Minsky finance-state classification, and an interactive Streamlit dashboard.
Investigated the viability of human-specialised neural processing as a basis for next-generation CAPTCHAs resilient to multimodal LLMs. Introduced ASCII art-based visual CAPTCHAs and noise-corrupted audio QA CAPTCHAs. No frontier model (GPT-4o, Gemini) solved a single ASCII CAPTCHA; all performed near-randomly on audio CAPTCHAs.
Research project for obstacle detection and avoidance on a DJI Tello drone using deep learning depth estimation models and image processing algorithms. Conducted under Prof Mir Feroskhan and Dr Wen Nuan.
Privacy-preserving group scheduling app that finds shared free time from Google Calendar without exposing private event details. Members see colour-coded availability overlays and can propose meeting times with real-time conflict warnings.
Chrome extension that detects scam posts on Instagram, Facebook, Twitter, and Reddit using a two-model ML pipeline - TF-IDF logistic regression on post text fused with a metadata classifier (follower counts, engagement ratios, account age). Injects a real-time risk badge directly into each post with a full analysis overlay.